In vivo percutaneous reflectance spectroscopy of fatty liver development in rats suggests that the elevation of the scattering power is an early indicator of hepatic steatosis

Daqing Piao*†¶, Jerry W. Ritchey, G. Reed Holyoak, Corey R. Wall, Nigar Sultana||, Jill K. Murray and Kenneth E. Bartels *School of Electrical and Computer Engineering Oklahoma State University, 202 Engineering South Stillwater, OK 74078, USA Department of Veterinary Clinical Sciences Center for Veterinary Health Sciences 002 VTH, Oklahoma State University Stillwater, OK 74078, USA Department of Veterinary Pathobiology Center for Veterinary Health Sciences Oklahoma State University, 250 McElroy Hall Stillwater, OK 74078, USA Graduate Program on Interdisciplinary Sciences Oklahoma State University, Stillwater, OK 74078, USA ¶daqing.piao@okstate.edu


Introduction
Fatty livers are used to address the severe donor organ shortage in liver transplantation 1 in the United States 2 and worldwide.Livers with mild level of steatosis, preferrably morphometrically microvesicular type of steatosis, 3 are considered for transplant.Microvesicular steatosis is not associated with primary graft nonfunction after transplantation. 4In contrast, macrovesicular steatosis is a known risk factor to ischmetic reperfusion injury 5,6 post transplantation and could cause adverse outcomes in living donors who undergo right hepatectomy. 7][5][6][7] Histopathological evaluation of liver biopsy specimens grades steatosis according to a semi-quantitative metrics 8 : a grade of \absent" for < 5% hepatocytes a®ected by lipid in¯ltration, \mild" for 5% to < 30% in¯ltration, \moderate" for 30% to < 60% in¯ltration, and \severe" for !60% in¯ltration.Gross assessment of the liver by visual inspection and palpation has shown low accuracy of identifying the parenchyma steatosis. 9,10Imaging modalities such as computed tomography and magnetic resonance imaging/spectroscopy provide highly accurate assessment of hepatic steatosis 11 but may be logistically un¯t for procurements.Ultrasound (US) imaging is the de facto modalidy for bedside evaluation of steatosis using speci¯c diagnostic markers including hepato-renal index. 12ltrasound imaging, however, lacks the sensitivity to reliably identify livers with mild level of steatosis. 13Alternative imaging or sensing tools have been investigated for potentially bedside real-time quanti¯cation of fatty content in liver tissue. 14Xu et al. 15 demonstrated photoacoustic spectrum analysis for di®erentiating fatty from normal livers.Other optical spectroscopic methods were applied in various forms and clinical operatabilities.Evers et al. 16 and Westerkamp et al. 17 used needle-based di®use re°ectance spectroscopy (DRS) over a broad visible-near-infrared spectrum up to 1600 nm for measuring within the liver parenchyma.It was shown that the strong lipid absorption around 1210 nm was a robust diagnostic marker for identifying hepatic steatosis in human liver specimen or during the surgery. 16Nilsson et al. 18 reported that surface DRS measurements from normal or tumorous liver represented the whole liver.A timeresolved near-infrared spectroscopy study by Kitai et al. 19 on rat liver specimens with di®erent levels of diet-induced fatty changes revealed that fatty liver presented lower absorption coe±cient ( a Þ and higher reduced scattering coe±cient ( 0 s Þ when compared with normal liver, suggesting the fat droplets inside the hepatocytes as the dominant scatterers over the visible to near-infrared spectrum.However, the robustness of detecting lipid deposition using the absorption of lipid that is prominent at 1210 nm degrades as the amount of lipid reduces, making it less sensitive or less speci¯c for detecting mild level of steatosis. 16,18here was also interest in whether the scattering information alone acquired over a spectral range not sensitive to lipid absorption could signify the onset of steatosis.Surface measurements on human liver specimens by McLaughlin et al. 20 using DRS over 550-1040 nm showed a correlation between the di®use re°ectance intensity and the histological lipid mass fraction of liver tissue.DRS has also shown to complement liver biopsy for grading hepatic ¯brosis in para±n-preserved human liver specimens. 21While the increased lipid content is expected to globally increase the di®use re°ectance of a fatty liver in comparison to a lean liver, 20 the onset of lipid droplets and the changes of the lipid droplet size shall also alter the spectral dependence of the scattering.Evers et al. 16 have already hypothesized that lipid in¯ltration changes the scattering powerthe indicator of the scattering spectral dependence.This hypothesis, however, remained unexamined until now. This study reports that the fatty changes of livers could cause the elevation of the scattering power, at the early stage of hepatic steatosis development even when the steatosis was unremarkable on ultrasonography.There is currently no alternative way to estimate the scattering power of a bulk tissue in vivo other than isolating the contribution of it to the scattering spectrum acquired by re°ectance spectroscopy. 22This study thus employed a minimally invasive approach of percutaneous single-¯ber spectroscopy (per-SfS) to allow direct sampling of the liver parenchyma during longitudinal evaluation of the same liver undergoing steatosis development.This study also used the histopathological imagery to analyze how the size of lipid droplets changed at di®erent levels of steatosis.The changes of the lipid size at the early stage of the steatosis provided morphometric evidence of the changes of the scattering power resolved from the per-SfS measurements.

Methods and Materials 2.1. Animals -Diets and timelines
The animal study was approved by the Institutional Animal Care and Use Committee of Oklahoma State University (protocol #VM-11-20).A total of 24 rats were separated into two phases (referred to as phase-I and phase-II, as shown in Fig. 1 were acclimated to laboratory conditions for a minimum of seven days when fed a standard rodent chow ad libitum (Laboratory Rodent Diet 5001, LabDiet, St Louis, MO, USA).The animals were allowed water ad libitum throughout the duration of the study.The 12 animals in each of the two phases were randomly divided with eight in the test group and four in the control group.The rats in the test group were fed a methionine-choline-de¯cient (MCD 23,24 ) diet (Harlan Telkad -TD.90262) ad libitum.The control rats were fed an amino acid control diet ad libitum (Harlan Telkad -TD.130936).The two phases when combined produced a sample size of 8 for the control group and 16 for the MCD-treated group with subgroups di®ering in size and the level of steatosis induced.After initiation of the respective diets, three rats including one of the control group and two of the MCD-diet fed group were euthanized at randomly set time intervals of, respectively, 12 (phase-I), 13 (phase-II), 27 (phase-II), 28 (phase-I), 41 (phase-II), 49 (phase-I), 55 (phase-II), and 77 (phase-I) days, for histological sampling of the livers in the respective groups of control and MCD-diet fed rats.Rats were euthanized after undergoing per-SfS measurements.In vivo per-SfS measurements presented in this work were those performed on all animals at the baseline, on day 12 of phase-I combined with day 13 of phase-II, and upon the removal of all animals for histological sampling.

Experimental procedures
The experimental procedures for per-SfS of the rat liver in vivo with ultrasound guidance of the ¯berprobe placement are illustrated in Fig. 2. The rat was surgically prepared under general anesthesia.After sonographic evaluation of the liver, a sterile, 22 or 20-gage 1.5 inch spinal needle (depending upon the availability) was inserted percutaneously into the liver under ultrasound guidance.After retracting the stylet of the needle, a sterile 320 m single-¯ber probe was inserted through the needle into liver parenchyma.The placement of the ¯ber probe into the parenchyma via the needle was monitored by ultrasound until the ¯ber tip extended a few millimeters beyond the beveled needle tip.Doppler ultrasound was also used to monitor that vasculature was not apparent in the vicinity of the ¯ber tip.After stabilizing the ¯ber placement, ¯ve repeated SfS measurements were acquired from each subject for o®line model-based processing of the spectral parameters including the scattering power.After per-SfS measurements, the rats slated to continue in the study were recovered and returned to their cages for observation.After per-SfS measurements, a rat to be euthanized for harvesting liver specimens underwent a ventral midline incision that extended cranially through the xiphoid and diaphragmatic re°ection into the thoracic cavity using a #10 scalpel blade and metzenbaum surgical scissors.A 20-gage needle with a 3 cc The experimental con¯guration for percutaneous SfS of rat liver.A broadband light source and a compact VIS/NIR spectrometer were coupled, respectively, to one ¯ber branch (200 m core) of a bifurcated ¯ber bundle.The combined terminal of the bifurcated ¯ber bundle (400 m diameter) was connected to a 320 m single-¯ber applicator probe.The single-¯ber probe was introduced into rat liver through a 22 (or 20) gage needle with ultrasound guidance.(b) A photograph of an anesthetized rat that underwent percutaneous SfS assessment of the liver with ultrasound monitoring of the ¯ber-probe placement.Marked in the photograph was also a small tubing connecting to a pneumatic pillow sensor placed at the left dorsal thoracic aspect of the rat for triggering SfS data acquisition at the same respiratory phase.
syringe was then used to collect $ 2 mL of whole blood by direct intracardiac puncture for biochemical analysis of plasma.Immediately following blood collection, direct intracardiac injection of 1 cc (390 mg) of pentobarbital sodium (Beuthansia D, Schering Plough, Union, New Jersey, USA) was administered for euthanasia.The body weight was measured immediately after euthanasia, followed by necropsy and harvesting of liver specimens.Tissue specimens were harvested from the right lobe of the excised liver by free-hand techniques.The specimens were harvested at arbitrary sites not correlated with the site of per-SfS, as it was not possible to accurately identify the site of in vivo per-SfS probing within the liver parenchyma when the liver was excised.The hepatic tissue specimens were ¯xed in 10% neutral bu®ered formalin, which does not cause false positive of lipid deposition in tissue. 25The liver specimens were then prepared for hematoxylin/eosin (H&E) staining and Oil-Red-O staining. 26The histology was examined on the H&E-stained and Oil-Red-O-stained specimens only.A total of eight H&E-stained specimens and four Oil-Red-O-stained specimens were examined for each liver.The H&Estained sections were cut at 4 m thickness.The Oil-Red-O-stained specimens were cut at 10 m thickness.A board-certi¯ed pathologist (J.W. R.) blinded to the groups ranked the H&E and Oil-Red-O (for con¯rmation of lipid)-stained specimens based upon the percentage amount of hepatocytes showing lipid accumulation.The identi¯cation of each animal (controlled by K. E. B.) was also blinded to the operators of per-SfS (D.P. and N. S.) as well as ultrasound (G.R. H. and C. R. W.) until the necropsy examinations of all animals were completed.

Biochemical analysis of plasma
Plasma samples were analyzed in the clinical laboratory of the veterinary teaching hospital of Oklahoma State University using standard laboratory methods.The analysis reported included the following serum markers: albumin, globulin, total bilirubin (T-Bil), glucose, cholesterol, triglyceride, sodium, chloride, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and alkaline phosphatase.

Spectral analysis speci¯c to per-SfS
The systems and methods of per-SfS had been detailed in other studies. 27The per-SfS system, including the broadband source, ¯ber-probe, and compact spectrometer, produced a working spectral response in a narrow 400 nm range over 540-940 nm, away from the Soret band of blood-related molecules such as porphyrins. 28The spectral analysis for per-SfS follows the semi-empirical model detailed by Kanick et al. [29][30][31][32] and others 33,34 corresponding to single-¯ber probing of a scattering medium by using a 15 angle-polished ¯ber.The SfS signal R tiss ðÞ, due exclusively to the tissue properties, after normalizing the raw signal using reference measurements, was 35 : where n eff is an e®ective index of scattering intensity formed when all isolated constant terms are combined, A is the scattering amplitude, b is the scattering power, HbO a ðÞ and HbR a ðÞ are, respectively, the absorption coe±cients of 1 M oxyhemoglobin and 1 M deoxyhemoglobin (values exported from the spectral panel of VirtualPhotonics 36 ), StO 2 is the hemoglobin oxygen saturation, ½HbT is the total hemoglobin concentration (in M), Lip a ðÞ and Wat a ðÞ are, respectively, the absorption coe±cients of lipid and water at 1% volume fraction (values also exported from the spectral panel of VirtualPhotonics 36 ), and f Lip is the fraction of lipid in the water-lipid body.The fraction of water and lipid in tissue was assumed a constant value of 93% following studies speci¯c to liver. 37The pigment packaging e®ect 38 was implemented but was found insensitive as the Soret band of blood-related molecules such as porphyrins was distant from the e®ective spectral response of the system.A nonlinear least-square ¯tting of the measurement data with Eq. ( 1) then estimated the parameters of interest.

Morphometric particle analysis on the histology images
The images of H&E-stained liver specimens were processed using ImageJ 39 for morphometric analysis of the area fraction, total count, and average size of the lipid droplets. 40The steps of the morphometric analysis are outlined in Fig. 3. On an original colored H&E image (Panel A), the lipid droplets are clearly visualized as circular or slightly elliptical particles.The original colored H&E image was ¯rst converted to an 8-bit gray-scale image (Panel B), which was then black-white inverted so the lipid droplets would appear black (Panel C).
The black-white inverted image was then applied an upper threshold (20 out of 255) of the gray scale to remove inter-hepatocyte structures not indicating lipid droplet features (Panel D), followed by particle analysis.All particles with the circularity between 0.5 and 1.0 and the diameter between 0.1 m and 50 m were counted (Panel E).The upper limit of the particle size was set at 50 m, after gross examination of the largest lipid droplets on all specimens, for avoiding automated counting of some large structures that were not lipid droplets (e.g., the central vein).The analyses produced particle size distribution (Panel F), from which the total count of the particles and the average particle diameter were calculated, and the area fraction (%) occupied by the counted particles.Similar procedures were tested with the Oil-Red-O images, but the particle analyses with the Oil-Red-O images were found not as robust as with the H&E images, due to the limited color contrast of the acquired Oil-Red-O images.The Oil-Red-O images were thus used for con¯rming the extensity of the lipid accumulation as assessed on the H&E images, but not for deriving the particle morphometry.

Statistical analysis
Statistical data analysis was performed with GraphPad Prism 6 (GraphPad Software, La Jolla, CA, USA).The groups compared had di®erent sample sizes.The control group had a sample size of 8.The test group had a sample size of 16 when all were combined, or a sample size of 7, 3, or 6 depending on the pathologic level of hepatic steatosis.Paired t-test was applied to the same group at two di®erent time-points to determine whether a speci¯c parameter of the group changed over time.Mann-Whitney test was applied to two groups of di®erent sample sizes to determine whether there was a di®erence of a speci¯c parameter between two groups at the same time-point.One-way analysis of variance (ANOVA) was applied to three or more groups of data.p < 0:05 infers statistically signi¯cant di®erence. 41The results were plotted as the mean value and the error bars that represent the standard deviation.

Ultrasound and per-SfS of representative levels of steatosis
The representative results including transabdominal ultrasound and per-SfS of the rat livers with di®erent levels of steatosis are presented in Fig. 4.
The top row corresponds to the baseline ultrasound images of the rats on day 0. The second row displays the ultrasound images at the day of euthanasia.The third row is for the per-SfS measurements on the day of euthanasia.The per-SfS result of a test rat is plotted against that of a control rat euthanized on the same day.Shown in the fourth row are the images of H&E staining and the bottom row images of Oil-Red-O staining.The column marked as \control" was from a control rat scari¯ed on day 13.No droplet structures indicating lipid accumulation in hepatocytes were found in the H&E-stained and Oil-Red-Ostained specimens.For this control rat, the sonographic features of the liver on day 13 appeared normal as was on day 0. The per-SfS of this control liver resolved a scattering power of 0.035.
The column marked as \mild" was from an MCD-diet-treated rat scari¯ed on day 13.Lipid in¯ltration in less than 30% of hepatocytes was found in the H&E-stained and Oil-Red-O-stained specimens, with an evident level of microvesicular steatosis.For this MCD-diet-treated rat with mild lipid accumulation, the sonographic features of the liver on day 13 were similar in echogenicity when compared with the day 0 livers.When comparing between the two per-SfS pro¯les, the spectral intensity of the one corresponding to the MCD-diettreated liver declined with the wavelength increasing from $ 700 nm.The per-SfS of this MCD-treated liver resolved a scattering power of 0.33 in comparison to its control counterpart of 0.035.
The column marked as \moderate" was from an MCD-diet-treated rat euthanized on day 27.Lipid in¯ltration in greater than 30% but less than 60% of hepatocytes was found in the H&E-stained and Oil-Red-O-stained specimens.Microvesicular steatosis was present in this specimen, but at a much lesser amount when compared with that in the mildly in¯ltrated specimen.For this MCD-diet-treated rat with moderate lipid accumulation, the sonographic features of the liver on day 27 showed di®use increased parenchymal echogenicity, development of hepatomegaly, some loss of visibility of hepatic vascular architecture and mild sound attenuation in the presence of echogenic liver, which together supported fatty liver diagnosis. 14When comparing between the two per-SfS pro¯les, the spectral intensity of the one corresponding to the MCDdiet-treated liver declined with the wavelength increasing from $ 700 nm.The per-SfS of this MCDdiet-treated liver resolved a scattering power of 0.31 in comparison to its control counterpart of À0:33.
The column marked as \severe" was from an MCD-diet-treated rat euthanized on day 55.Lipid in¯ltration in greater than 60% of hepatocytes was found in the H&E-stained and Oil-Red-O-stained specimens.For this MCD-diet-treated rat with severe lipid accumulation, the sonographic features of the liver on day 55 presented marked sound attenuation leading to poor visualization of the diaphragm and hepatic vessels in addition to highly di®use increased parenchymal echogenicity, which were diagnostic of fatty in¯ltration. 16When comparing the per-SfS pro¯le between the control rat and the MCD-diet-treated rat, the spectral intensity of the liver of the MCD-diet-treated rat was noticeably higher than that of the control rat, and the spectral intensity of it declined with the wavelength increasing from $ 700 nm.The per-SfS of this MCD-diet-treated liver resolved a scattering power of 0.55 in comparison to its control counterpart of 0.21.Column (1): The one marked as \control" below the H&E image was from a control rat of phase-II that was sacri¯ced on day 13.Column (2): The one marked as \mild" below the H&E image was from an MCD-diet-treated rat in phase-II that was sacri¯ced on day 13 (the same set of euthanasia including the control rat shown here as \control").Column (3): The one marked as \moderate" below the H&E image was from an MCD-diet-treated rat in phase-II that was sacri¯ced on day 27.Column (4): The one marked as \severe" below the H&E image was from an MCD-diet-treated rat of phase-II that was sacri¯ced on day 55.Dimension of the bar on the histology image ¼ 100 m.

Lipid morphometry corresponding to the grade of steatosis and the length of diet intake
The results of particle analysis on liver specimens absent of and presented with mild, moderate, and severe levels of steatosis are given in Fig. 6.The particle size distributions are plotted at the same scales, for particle diameters up to 30 m and particle count up to 2000.The particle count is displayed at a logarithmic scale to magnify the changes occurring as larger droplets appearing at higher grade of steatosis.The dashed arrows on the ¯gure indicate the same plot of the particle size distribution over its subsequent ¯gure to be used as a reference to the particle sizes of the current specimen.
When comparing the specimen of mild steatosis with the control one, the number of droplet of sub-m size increased substantially and droplets of $ 10 m size started to appear.When comparing the specimen of moderate steatosis with the one of mild steatosis, the number of droplet structures less than 8 m in diameter decreased, but the number of droplet structures greater than 8 m increased sig-ni¯cantly, with the appearing of droplets as large as 20 m.When comparing the specimen of severe steatosis with the one of moderate steatosis, the number of droplet structures less than 8 m in diameter decreased, but the number of droplet structures greater than 8 m increased signi¯cantly, with the appearing of droplets as large as 30 m.
The morphometric results obtained from the H&E images of all 24 rats (8 control and 16 MCDdiet-treated) are displayed in Figs.7(a) to 7(c) with respect to the terminal duration (in weeks) of each animal on its respective diet: (a) is the area fraction, (b) the total droplet count, and (c) the mean diameter of the droplet structures.The same Fig. 6.The top row shows the images of the H&E-stained specimens at four levels of steatosis: control, mild, moderate, and severe.These four images are identical to those on row 4 of Fig. 4. The images at the middle row are the images at the top row after morphometric analysis to mark the lipid droplets, equivalent to (e) of Fig. 3.The bottom row displays the particle size distribution.X-axis: particle diameter (m).Y -axis: particle count.The change from control to mild is presented with the appearance of primarily small particles (diameter < 10 m, the particle count increases logarithmically at smaller particle).The change from mild to moderate is presented with the appearance of particles of 10-20 m.The change from moderate to severe is presented with continued increase of particles of 10-20 m and the appearance of particles of sizes up to 30 m. Dimension of the bar on the histology image ¼ 100 m. morphometric analyses results are displayed in Figs.7(a 0 Þ to 7(c 0 Þ, with respect to the histopathologically assessed grade of the steatosis.The morphometric area fraction of the droplet structures in the control specimens (n ¼ 8) ranged from 0.14% to 2.22%.The morphometric area fraction of the droplet structures in the mild MCD-diet-treated subset (n ¼ 7) ranged from 2.12% to 9.76%.The morphometric area fraction of the droplet structures in the moderate MCD-diet-treated subset (n ¼ 3) ranged from 5.90% to 16.69%.The morphometric area fraction of the droplet structures in the severe MCD-diet-treated subset (n ¼ 6) ranged from 23.14% to 36.14%.The morphometric area fraction of the MCD-diet-treated livers, when plotted with respect to the length of diet intake as shown in (a), has shown to be generally higher if treated longer with the MCD diet up to 8 weeks.The area factions of the two rats examined at the 11 weeks were, however, signi¯cantly lower than with respect to the duration on the diet and a clear bi-phasic pattern in (b 0 ) with respect to the pathological grade of the steatosis.The total droplet count of specimens was substantially greater (more than two folds on average) in the specimens with mild steatosis when compared with the control specimens, but were less in the moderate steatosis when compared with the mild steatosis, and less in the severe steatosis when compared with the moderate steatosis.When the total droplet count of the MCD-diet-treated specimens was plotted against the morphometric area fraction as is shown in Fig. 8 (a), the total count was shown to be inversely proportional to the area fraction, agreeing with other reports for MCD-diet-induced hepatic steatosis in mice. 42When the mean droplet diameter of the MCD-treated specimens was plotted against the morphometric area fraction as shown in (b), a linearity of R ¼ 0:93 between the size and the area was established, indicating that higher steatosis grade is associated with bigger droplet size.

Scattering power with respect to the grade of steatosis
The distribution of the terminal scattering powers resolved by per-SfS for the 24 rats is presented in Fig. 9 The scattering powers of the 24 rats on the day 12/13 timeline of the two phases combined were also compared in (d) with the label of \Day 12/13".For these 24 rats on the day 12/13 timeline, the pathological steatosis grades of only six livers (two control and four MCD-diet treated) were known.On day 12/13, the scattering power of the livers in the test group (0:32 AE 0:17, n ¼ 16) was signi¯cantly elevated than (p < 0:0017) that of the control livers (0:10 AE 0:11, n ¼ 8).The scattering powers of the 24 rats on the day of initiating their respective diets were also compared in (d) with the label of \Day 0".On day 0, the scattering power of the livers in the test group (0:09 AE 0:19, n ¼ 16) and that of the control livers (0:10 AE 0:17, n ¼ 8) were similar (p ¼ 0:93).The change of the scattering power of the control group from day 0 to day 12 was not evident (p ¼ 0:22) but that of the test group was signi¯cant (p < 0:0001).The change of the scattering power of the control group from day 0 to day of euthanasia was not evident (p ¼ 0:23) but that of the test group was signi¯cant (p ¼ 0:0029).

Other outputs of the spectral analysis
The total hemoglobin concentrations of 8 control rats and 16 test rats on day 0 were 441:6 AE 48:0 M and 410:9 AE 50:0:5 M, respectively (p ¼ 0:83).The total hemoglobin concentrations of the 8 control rats and 16 test rats on the day of euthanasia were 458:8 AE 79:1 M and 334:3 AE 53:8 M, respectively (p ¼ 0:14).The reduction of the total hemoglobin of the 16 test rats on the day of euthanasia in comparison to their respective baselines on day 0 was not signi¯cant (p ¼ 0:30).No signi¯cant di®erences were found for the hemoglobin oxygenation when compared longitudinally within the same group or laterally between the control and MCD-diet-treated group.No signi¯cant di®erences were found for the e®ective scattering intensity when compared longitudinally within the same group or laterally between the control and MCD-diet-treated group.In summary, no signi¯cant di®erence was found in parameters other than the scattering power that were retrieved from the per-SfS measurements, including total hemoglobin, hemoglobin oxygenation, water, lipid fraction, and e®ective scattering intensity, between the MCD-diet-treated group and the control group at the same time-point or between two time-points of the same group.

Biochemical parameters
Due to the missing of data log, the biochemical parameters were incomplete.Records of albumin, globulin, total bilirubin, glucose, cholesterol, sodium, chloride, AST, ALT, and ALK phosphatase were retrieved for 6 out of 8 rats in the control group and 12 out of 16 rats in the MCD-diet-treated group.Records of triglyceride were retrieved for 5 out of 8 rats in the control group and 10 out of 16 rats in the MCD-diet-treated group.The results are presented in Fig. 10, which also informs the body weight.Records of body weight upon necropsy were retrieved for 4 out of 8 rats in the control group and 8 out of 16 rats in the MCD-diet-treated group.Plasma albumin was slightly increased in the MCD-fed rats with statistical signi¯cance (p ¼ 0:0027).No change of plasma globulin was seen in the MCD-diet-treated rats.Plasma albumin was slightly increased in the MCD-fed rats with statistical sig-ni¯cance (p ¼ 0:0027).No changes of plasma glucose and cholesterol were seen in the MCD-diet-treated rats.Plasma triglyceride was reduced in the MCDfed rats with statistical insigni¯cance (p ¼ 0:27).No changes of plasma sodium and chloride contents were seen in the MCD-diet-treated rats.Plasma AST was increased in the MCD-fed rats with statistical insig-ni¯cance (p ¼ 0:12).Plasma ALT was increased to nearly two folds in the MCD-fed rats (p ¼ 0:0097).No change of plasma ALK phosphatase was seen in the MCD-fed rats.In addition to the biochemical parameters, the body weight was reduced in the MCD-fed rats (p ¼ 0:0040).

Discussion
The scattering power resolved for the control livers is 0.1, a number that is in close agreement with the number reported for liver by van Leeuwen-van Zaane et al. 34 Among the 16 rats fed MCD diet, 7 had mild lipid accumulation, 3 had moderate lipid accumulation, and 6 had severe lipid accumulation.The scattering powers of the livers with pathological steatosis were all substantially greater than those of the control livers.Within the MCD-diet-treated group, the scattering power was higher in moderate than in mild subsets, but lower in the severe than in the moderate subsets.The pattern of change of the scattering power was projected to be caused by an initial accumulation of primarily small lipid droplets prior to the lipid droplets increasing in size. 35In the previous report of re°ectance spectroscopy, 35 it was demonstrated in simulation that an increase of 1 m lipid droplet from 1% to 15% concentration in a lean liver could increase the scattering power from 0.1 of the lean liver to 0.13 and 0.31, and an increased lipid size from 2 m to 25 m at the same 33% volume concentration could cause the scattering power to decrease from 0.30 to 0.19.Such postulated changes of the lipid droplets are in agreement with the global changes of the lipid droplets as analyzed on the histopathological images in this present study containing both phases of the study.As shown in Fig. 6, the change from lean liver to liver with mild steatosis was associated with a substantial increase of smaller sub-m size lipid droplets.As the steatosis elevated in grade, the number of smaller droplets decreased, but larger droplets with diameters ranging from 8 m to 20-30 m increased signi¯cantly.These morphometrically analyzed changes of the lipid droplet size also agreed with the patterns reported by Veteläinen et al. 43 from similar rat modeling work.
It is shown that the morphometric area fraction determined by the automated image analysis correlates well with the visual estimation of the liver steatosis by a pathologist, but the area fraction determined by the automated image analysis is systematically lower than that by the pathologist's visual estimation of the amount of number of hepatocytes with the presence of lipid in¯ltration.For example, on evaluation of specimens which had pathologically severe ð> 60%Þ steatosis (n ¼ 6), the automated morphometric analysis returned an area fraction ranging between 23.14% and 36.14%.
On evaluation of specimens which had pathologically moderate (30%-60%) steatosis (n ¼ 3), the automated morphometric analysis returned an area fraction ranging between 5.90% and 16.69%.On evaluation of specimens which had pathologically mild ð< 30%Þ steatosis (n ¼ 7), the automated morphometric analysis returned an area fraction ranging between 2.12% and 9.76%.On evaluation of specimens which were pathologically absent ð< 5%Þ of steatosis (n ¼ 8), the automated morphometric analysis returned an area fraction ranging between 0.14% and 2.22%.Similar discrepancies between the objectively identi¯ed and subjectively estimated lipid area fractions were reported by Marsman et al. 44 who compared the average percentage of fat content in 49 biopsy specimens analyzed by automated software and a trained pathologist.It was reported that the macrovesicular fat content of donor liver specimens determined by the pathologist was 2.9-fold greater, on average, than that obtained by the automated software. 44Similarly, the pathologist's determination of total fat content was 1.4-fold greater, on average, than that obtained by the automated software. 44The di®erence between the visual estimation of steatosis grade and automated quantitation of the fat fraction is caused in part by the di®erence between how a pathologist evaluates and how an automated image analysis performs.A pathologist determines the steatosis grade based on the subjective evaluation of the fraction of the hepatocytes in¯ltrated by lipid, whereas what an automated software toolbox calculates is the area fraction of the lipid droplets over the total image area.The area fraction extracted from a two-dimensional image by using the automated image analysis package was used directly in this study for keeping it consistent with the subjective estimation by the pathologist of the number of hepatocytes a®ected on a two-dimensional image.Should the lipid fraction be represented as the volume fraction, 16,44 the percentage values of the lipid content could have been even smaller.But the absolute values of the lipid fraction will not have affected the proportionality that it has with respect to either the lipid droplet size or count.
Several limitations are noted in this study that replied upon longitudinal assessments of the rat livers in vivo using per-SfS to identify a potential spectroscopic marker of steatosis development.One concern was that liver specimens obtained from three rats (two MCD and one control) might not represent histological sampling for the whole experimental group at a given time-point.This was particularly concerning for sampling the experimental groups several weeks after the diet induction when the randomly selected animals could have the highest or lowest level of steatosis among the group at the stage of experiment.However, the undersampling of the histology at a given time-point did NOT a®ect the evaluation of the histology of the whole group at the respective end-points (euthanasia).The under-sampling of the histology at a given time-point could have a®ected the evaluation at day 12/13 when 6 rats out of 24 rats (or 4 out of 18 MCD-diet-treated rats) were sampled.For the evaluation on day 12/13, the duration of the diet intake was relatively short at the near-2 weeks timepoint and mild lipid in¯ltration was found in all of the four MCD-diet-fed rats.Therefore, it was a reasonable projection that all of the 16 MCD-dietfed rats would have had mild lipid in¯ltration at that time-point, which, however, could not be con-¯rmed.We note that there is no easy way both to accurately sample the histology of the (small) rat livers of the whole group and to longitudinally study the changes expected for the whole group over time.
0][31][32][33][34][35] The discrepancy between the model geometry and the actual geometry could a®ect the spectral analysis of the intensity values, even though it may not change the relative pattern of the scattering spectra that is eventually dictated by the di®erence in the scattering powers.Thirdly, percutaneous ¯ber application could have caused injuries to the liver and confounded comparison at di®erent time-points.However, artifacts due to the injury would occur in both control and test groups; therefore, it should not adversely a®ect the comparison between the control and test groups at the same time-points (i.e., day 12/13).Fourthly, the transabdominal evaluation of the fatty liver conditions with sonography was based on the gray-scale features only; so the steatosis diagnosis was likely not as sensitive as would have been obtained with other diagnostic imaging markers such as using the hepato-renal index. 12,13Regardless of the diagnostic method used, the low sensitivity of ultrasonography in detecting a mild level of steatosis is however not unexpected.The lipid droplets formed during the initial stage of fatty in¯ltration could be one to two orders smaller than the ultrasonic wavlength in tissue.At a small lipid fraction of the total tissue during the onset of steatosis, the dispersed amount of the small lipid droplets could not change the acoustic impedance of the tissue su±ciently to increase the echogenicity.Lastly, the sampling of liver tissue in this study was done in a minimally invasive approach that would be less applicable than a noninvasive approach for clinical translation.The objective of this study was to discover a spectral feature that represented the \authentic" conditions of the liver over the course of diet intake, which was di±cult to assess by noninvasive optical spectroscopy methods.It should be noted that it is possible to use surface-based re°ectance spectroscopy applied directly with transabdominal re°ectance spectroscopy to assess the scattering spectral feature, but that requires additional modeling or device technologies to compensate the e®ect of cutaneous and subcutaneous tissue by using methods such as multiple source-detector pairs 45 for the DRS.
A lasting limitation involving MCD-diet model is that MCD-diet-induced fatty in¯ltration in liver does not represent accurately the biology of human nonalcoholic fatty liver disease (NAFLD). 46Although this model could replicate the histological features of liver injury as observed in human NAFLD, its metabolic context is distinct from human NAFLD, since animals fed the MCD diet lost weight [Fig.10(c)] and presented decreased blood triglyceride and cholesterol, creating a metabolic pro¯le opposite to the human disease.The main advantages of the MCD diet are that it is widely available and reliably replicates NAFLD histology within a relatively shorter feeding time than other dietary models of NAFLD.The histological relevance of MCD diet induction in terms of the fatty in¯ltration in liver has made it possible for this study to observe the changes of the scattering power and to associate the changes of the scattering power with the changes in the morphology of the lipid droplets in the liver.

Conclusions
In vivo per-SfS was performed on livers of rats fed MCD diet in comparison to control livers over randomized intervals.The rats in the MCD-diettreated group developed various levels of steatosis (seven of mild, three of moderate, and six of severe), relative to none in the control group (n ¼ 8).Images of hematoxylin & eosin-stained specimens were also analyzed morphometrically to extract the area fraction, total count, and mean size of the lipid droplet structures.The mean droplet size increased linearly with the increase of the lipid area fraction, but the total count followed a bi-phasic pattern with the increase of the lipid area fraction.The per-SfS-resolved scattering power for the MCD-diettreated livers (0:33 AE 0:21, n ¼ 16) was signi¯cantly ðp < 0:0189Þ greater than that for the control livers (0:036 AE 0:25, n ¼ 8).When measured at near-2 weeks (day 12/13) with none of the MCDdiet-treated livers presenting steatosis-diagnostic patterns on sonography but all four specimens of the MCD-diet-treated group had pathological mild lipid in¯ltration with evident level of microvesicular steatosis, the per-SfS-resolved scattering power of the MCD-diet-treated livers (0:32 AE 0:17, n ¼ 16) was signi¯cantly (p ¼ 0.0017) greater than that of the control livers (0:10 AE 0:11, n ¼ 8).The elevation of scattering power is shown to indicate the onset of steatosis in rat livers when the steatosis was not detected by ultrasound.
Fig. 1.The timelines of the two phases of the animal study.(a) In phase-I, the histopathology results of the livers were obtained from one control rat and two MCD-diet-treated rats after the in vivo per-SfS measurements, on day 12, day 28, day 49, and day 77, respectively.(b) In phase-II, histopathology results of the livers were obtained from one control rat and two MCD-diet-treated rats after the in vivo per-SfS measurements, on day 13, day 27, day 41, and day 55, respectively.X-axis: number of days on the respective diet.Y -axis: number of animals available on the day for in vivo examination.The curling arrow indicates removal by the end of the study on the day for harvesting the liver specimens.

Fig. 2 .
Fig. 2. (a)The experimental con¯guration for percutaneous SfS of rat liver.A broadband light source and a compact VIS/NIR spectrometer were coupled, respectively, to one ¯ber branch (200 m core) of a bifurcated ¯ber bundle.The combined terminal of the bifurcated ¯ber bundle (400 m diameter) was connected to a 320 m single-¯ber applicator probe.The single-¯ber probe was introduced into rat liver through a 22 (or 20) gage needle with ultrasound guidance.(b) A photograph of an anesthetized rat that underwent percutaneous SfS assessment of the liver with ultrasound monitoring of the ¯ber-probe placement.Marked in the photograph was also a small tubing connecting to a pneumatic pillow sensor placed at the left dorsal thoracic aspect of the rat for triggering SfS data acquisition at the same respiratory phase.

Fig. 3 .
Fig. 3. Sequence of image processing by ImageJ for morphometric analysis of the lipid droplets.(a) The original colored image of H&E-stained liver specimen.(b) An 8-bit conversion of the colored image.(c) The black-white inverted image of the 8-bit image.(d) An upper threshold of the gray scale applied to the black-white image.(e) The counted particles are overlapped on the threshold applied image.(f) An example of the particle size histogram as the output of the particle analysis (X-axis: droplet diameter in m; Y -axis: droplet count).Dimension of the bar on the histology image ¼ 100 m.

Fig. 4 .
Fig. 4. Rows are counted from the top.Row (1): ultrasonography at the baseline.Row (2): ultrasonography at the day of euthanasia.Row (3): SfS pro¯le taken at the day of euthanasia.X-axis: wavelength (nm); Y -axis: spectral intensity (arbitrary unit).Row (4): image of H&E-stained specimen.Row (5): image of Oil-Red-O-stained specimen.Columns are counted from the left.Column (1): The one marked as \control" below the H&E image was from a control rat of phase-II that was sacri¯ced on day 13.Column (2): The one marked as \mild" below the H&E image was from an MCD-diet-treated rat in phase-II that was sacri¯ced on day 13 (the same set of euthanasia including the control rat shown here as \control").Column (3): The one marked as \moderate" below the H&E image was from an MCD-diet-treated rat in phase-II that was sacri¯ced on day 27.Column (4): The one marked as \severe" below the H&E image was from an MCD-diet-treated rat of phase-II that was sacri¯ced on day 55.Dimension of the bar on the histology image ¼ 100 m.
The pathologist's grading of the steatosis of the 24 livers (8 control and 16 MCD-diet-treated) after assessing both H&E-and Oil-Red-O-stained specimens is summarized in Fig.5.The summary plot of the grading is also presented with representative Oil-Red-O images of the two MCD-diet-treated livers harvested on the near-2 week, near-4 week, near-6 week, near-8 week, and 11-week time-points of the study.All eight control livers were absent of lipid accumulation, regardless of the length on the control diet.All four MCD-diet-treated livers that were harvested at near-2 weeks presented mild lipid accumulation in the liver with various but evident amounts of microvesicular steatosis.Among the four MCD-diet-treated livers that were harvested at near-4 weeks, one presented mild lipid accumulation, two presented moderate lipid accumulation, and one presented severe lipid accumulation.Both of the two MCD-diet-treated livers that were harvested at near-6 weeks presented severe lipid accumulation.The two MCD-diettreated livers that were harvested at 7 weeks presented mild lipid accumulation in one and severe lipid accumulation in the other.Both of the two MCD-diet-treated livers that were harvested at near-8 weeks presented severe lipid accumulation.The two MCD-treated livers that were harvested at 11 weeks presented mild lipid accumulation in one and moderate lipid accumulation in the other.The rats fed MCD diet did not exhibit any other signi¯cant hepatic pathology outside of what is reported here, i.e., primarily the fatty accumulation within the hepatocytes.There was neither signi¯cant in°ammation nor ¯brosis.

Fig. 5 . 9 J
Fig. 5. Histological grading of the hepatic steatosis of the rats at di®erent lengths of the diet intake.X-axis: length in days of diet intake; Y -axis: severity of steatosis.Two specimens were harvested at $ 2, $ 4, $ 6, $ 8, and 11 weeks, respectively.The upper panel with horizontal strips illustrates the severity of steatosis by color (three levels of the darkness, increased darkness for mild, moderate, and severe) or font size (three sizes of the font, increased font size for mild, moderate, and severe), versus the length of diet induction up to 77 days or 11 weeks.The lower panel shows two images of Oil-Red-O-stained liver specimens harvested, respectively, from two rats identi¯ed at the top panel that presented the level of steatosis marked by the red closed line.Dimension of the bar on the histology image ¼ 100 m.

Fig. 7 .
Fig. 7. (a) Area fraction with respect to the duration of the animal on the respective diet.(a 0 Þ Area fraction with respect to the pathological grade of steatosis.(b) Total count of the lipid droplet structures with respect to the duration of the animal on the respective diet.(b 0 Þ Total count of the lipid droplet structures with respect to the pathological grade of steatosis.(c) Average droplet size with respect to the duration of the animal on the respective diet.(c 0 Þ Average droplet size with respect to the pathological grade of steatosis.In all sub-¯gures, the green square represents control rat, and the red circle represents MCDdiet-treated rat. Fig. 8. (a) Total count of the lipid droplet structures with respect to the area fraction.(b) Average droplet size with respect to the area fraction.In all sub-¯gures, the green square represents control rat, and the red circle represents MCD-diet-treated rat.In both ¯gures, a hand-sketched broken line is used to illustrate the global pattern.

Fig. 9 .
Fig. 9. (a) Distribution of the scattering power with respect to the pathological grade of steatosis.(b) The distribution of (a) presented as a bar-chart for the four subsets.(c) The ROC analysis results in an area-under-curve of 0.797 for using the scattering power to make the diagnosis of steatosis.(d) The scattering powers of the 8 rats in the control group and 16 in the MCD-diet-treated group, when evaluated on the day of initiating the respective diets, day 12/13 of two phases combined, and the day of euthanasia (bar charts: mean þ standard deviation).